-
Notifications
You must be signed in to change notification settings - Fork 3
/
featureExtractor.py
265 lines (254 loc) · 14.2 KB
/
featureExtractor.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
import warnings
warnings.filterwarnings("ignore")
from pyampd.ampd import find_peaks
import numpy as np
import scipy
from biosppy.signals import bvp
from scipy import io
from hrvanalysis import get_time_domain_features
from hrvanalysis import get_frequency_domain_features
import mat73
# properties
freq = 125
debug = False
part_id = 1
save_name = "features_{}.csv".format(part_id)
# Address of data
data_path = "data"
# Loading MIMIC II Dataset
try:
data_mat = mat73.loadmat(data_path + '/Part_{}.mat'.format(part_id))
if part_id==1:
data = data_mat['Part_1']
if part_id==2:
data = data_mat['Part_2']
if part_id==3:
data = data_mat['Part_3']
if part_id==4:
data = data_mat['Part_4']
except:
data_mat = io.loadmat(data_path + '/Part_{}.mat'.format(part_id))
if part_id==1:
data = data_mat['Part_1'][0]
if part_id==2:
data = data_mat['Part_2'][0]
if part_id==3:
data = data_mat['Part_3'][0]
if part_id==4:
data = data_mat['Part_4'][0]
# or Loading the sample dataset
# with open('data_sample_BP_PPG.pkl', 'rb') as f:
# ... [data] = pickle.load(f)
# Read all of subjects signals
signals = []
details = []
all_features = []
data_length = len(data)
for subject_id in range(0,data_length):
print('subject_id={}'.format(subject_id))
[PPG, BP, ECG] = data[subject_id]
# PPG = PPG[1:int(len(PPG))]
subject_details = [0,1,2,3,np.max(BP), np.min(BP), 6,7,8]
details.append(subject_details)
signals.append(PPG)
# Extracting BP-related features of each signal
for i,PPG in enumerate(signals):
try:
print("signal = {}".format(i))
# Filtering the signal
# the library uses filtfilt (=> Phase = 0)
ppg_filtered_low = bvp.st.filter_signal(signal=PPG,ftype='butter',band='lowpass',order=6,frequency=10,sampling_rate=freq)[0]
ppg_filtered = ppg_filtered_low - scipy.signal.medfilt(ppg_filtered_low, int(15 * 1000/freq)-1)
# Calculating the peaks and minimums of the signal (using AMPD)
onsets = find_peaks(-1 * ppg_filtered)
peaks = find_peaks(ppg_filtered)
nn_intervals_list = np.diff(onsets) * (1000/freq)
time_domain_features = list(get_time_domain_features(nn_intervals_list).values())
freq_domain_features = list(get_frequency_domain_features(nn_intervals_list).values())
HR = 60/(np.median(np.diff(onsets))/freq)
PPG = np.array(PPG)
subject_features = []
for onset in list(range(1,len(onsets)-1,5)):
try:
print(".", end="")
segment_features = []
for win_num in range(0,5):
ppg_window = ppg_filtered[onsets[onset + win_num]:onsets[onset + win_num + 1]]
ppg_window = ppg_window - np.min(ppg_window)
ppg_window = ppg_window / np.max(ppg_window)
window_features = []
ignore_flag = False
# Detecting maximum of the window and dividing the Pulse into two ascending and descending parts
ppg_window_systolic = np.argmax(ppg_window)
ppg_window_ascending = ppg_window[:ppg_window_systolic]
ppg_window_descending = ppg_window[ppg_window_systolic:]
# Fitting polynomials to each part and calculating their first and second derivatives
ppg_window_ascending_x = np.linspace(0,ppg_window_systolic,len(ppg_window_ascending))
ppg_window_descending_x = np.linspace(ppg_window_systolic,len(ppg_window),len(ppg_window_descending))
asc_weights = np.ones(np.shape(ppg_window_ascending))
asc_weights[int(0.25*len(asc_weights)):int(0.75*len(asc_weights))] = 1
asc_weights[int(0.4*len(asc_weights)):int(0.6*len(asc_weights))] = 1
des_weights = np.ones(np.shape(ppg_window_descending))
des_weights[int(0.25*len(des_weights)):int(0.75*len(des_weights))] = 1
des_weights[int(0.4*len(des_weights)):int(0.6*len(des_weights))] = 1
des_weights[int(0.8*len(des_weights)):int(1*len(des_weights))] = 1
ppg_window_ascending_poly5 = np.polyfit(ppg_window_ascending_x ,ppg_window_ascending,5,w=asc_weights)
ppg_window_descending_poly7 = np.polyfit(ppg_window_descending_x,ppg_window_descending,7,w=des_weights)
ppg_window_ascending_poly5_der = np.poly1d(np.polyder(ppg_window_ascending_poly5))
ppg_window_descending_poly7_der = np.poly1d(np.polyder(ppg_window_descending_poly7))
ppg_window_descending_poly7_der2 = np.poly1d(np.polyder(ppg_window_descending_poly7,2))
ppg_window_ascending_poly5_der_values = ppg_window_ascending_poly5_der(ppg_window_ascending_x)
ppg_window_descending_poly7_der_values = ppg_window_descending_poly7_der(ppg_window_descending_x)
ppg_window_descending_poly7_der2_values = ppg_window_descending_poly7_der2(ppg_window_descending_x)
# calculating the location of Max. slope point
ppg_maxslope = np.argmax(ppg_window_ascending_poly5_der_values)
# calculating the location of Diastolic peak, Dicrotic Notch and inflection point based on the method presented in our paper
ppg_window_descending_poly7_der_roots = np.sort(np.array([np.real(root) for root in np.roots(ppg_window_descending_poly7_der) if ~np.iscomplex(root)]))
ppg_window_descending_poly7_der_roots = np.array([root for root in ppg_window_descending_poly7_der_roots if (root>ppg_window_systolic*1.25 and root<0.95*len(ppg_window))])
ppg_window_descending_poly7_der_roots_der2_values = ppg_window_descending_poly7_der2(ppg_window_descending_poly7_der_roots)
if len(ppg_window_descending_poly7_der_roots)==0:
ppg_no_dia = True
else:
dia_candidates = []
for dia_index, dia_candidate in enumerate(ppg_window_descending_poly7_der_roots):
if ppg_window_descending_poly7_der_roots_der2_values[dia_index]<0:
dia_candidates.append(dia_candidate)
if len(dia_candidates)>0:
ppg_dia = dia_candidates[int(len(dia_candidates)/2)]
ppg_no_dia = False
else:
ppg_no_dia = True
if ppg_no_dia == True:
ppg_des_der2_minimums = scipy.signal.find_peaks(-1*ppg_window_descending_poly7_der2_values)[0] + ppg_window_systolic
dia_candidates = np.array([root for root in ppg_des_der2_minimums if (root>ppg_window_systolic*1.25 and root<0.95*len(ppg_window))])
if len(dia_candidates)>0:
ppg_dia = dia_candidates[0]
else:
ppg_dia = ppg_window_systolic + int(len(ppg_window_descending)/2)
# find dicrotic notch
ppg_des_der2_maximums = scipy.signal.find_peaks(ppg_window_descending_poly7_der2_values)[0] + ppg_window_systolic
dicrotic_candidates = np.array([root for root in ppg_des_der2_maximums if (root>ppg_window_systolic*1.2 and root<0.995*ppg_dia)])
if len(dicrotic_candidates)>0:
ppg_dic = dicrotic_candidates[0]
else:
ppg_dic = ppg_des_der2_maximums[0]
# find inflection point
ppg_window_descending_poly7_der2_roots = np.sort(np.array([np.real(root) for root in np.roots(ppg_window_descending_poly7_der2) if ~np.iscomplex(root)]))
inf_candidates = np.array([root for root in ppg_window_descending_poly7_der2_roots if (root>ppg_dic and root<ppg_dia)])
if len(inf_candidates)>0:
ppg_inf = inf_candidates[0]
else:
ignore_flag = True
ppg_inf = (ppg_dia + ppg_dic)/2
ppg_dia = int(ppg_dia)
ppg_dic = int(ppg_dic)
ppg_inf = int(ppg_inf)
# When point detection is done, extract the features
window_features.extend(details[i])
meanBP = (1*details[i][4] + 2*details[i][5]) / 3
window_features.extend([0])
window_features.extend([meanBP])
ppg_window_main = PPG[onsets[onset + win_num]:onsets[onset + win_num + 1]]
ppg_window_main = np.array(ppg_window_main)
ppg_dc = np.mean(PPG)
ppg_window_main_rev = ppg_window_main[::-1]
try:
ppg_last_min = len(ppg_window_main)- scipy.signal.find_peaks(-1*ppg_window_main_rev)[0][0]
except:
ppg_last_min = len(ppg_window_main) -1
ppg_last_min = (ppg_window_main[0]+ppg_window_main[ppg_last_min])/2
#
ppg_ac = np.median(PPG[peaks]) - np.median(PPG[onsets])
# modified Normalized Pulse Volume
mNPV1 = ppg_ac / (ppg_dc)
mNPV2 = ppg_ac / (ppg_dc+ppg_ac)
# Area related features
A1 = np.trapz(ppg_window_main[0:ppg_maxslope])
A2 = np.trapz(ppg_window_main[ppg_maxslope:ppg_window_systolic])
A3 = np.trapz(ppg_window_main[ppg_window_systolic:ppg_dic])
A4 = np.trapz(ppg_window_main[ppg_dic:ppg_inf])
A5 = np.trapz(ppg_window_main[ppg_inf:ppg_dia])
A6 = np.trapz(ppg_window_main)
S1 = A1/A2
S2 = A1/A3
S3 = A1/A4
S4 = A1/A5
S5 = A1/A6
S6 = A2/A3
S7 = A2/A4
S8 = A2/A5
S9 = A2/A6
S10 = A3/A4
S11 = A3/A5
S12 = A3/A6
S13 = A4/A5
S14 = A4/A6
S15 = A5/A6
S16 = (A1+A2+A3+A4)/(A5+A6)
#
NA1 = np.trapz(ppg_window[0:ppg_maxslope])
NA2 = np.trapz(ppg_window[ppg_maxslope:ppg_window_systolic])
NA3 = np.trapz(ppg_window[ppg_window_systolic:ppg_dic])
NA4 = np.trapz(ppg_window[ppg_dic:ppg_inf])
NA5 = np.trapz(ppg_window[ppg_inf:ppg_dia])
NA6 = np.trapz(ppg_window)
NS1 = NA1/NA2
NS2 = NA1/NA3
NS3 = NA1/NA4
NS4 = NA1/NA5
NS5 = NA1/NA6
NS6 = NA2/NA3
NS7 = NA2/NA4
NS8 = NA2/NA5
NS9 = NA2/NA6
NS10 = NA3/NA4
NS11 = NA3/NA5
NS12 = NA3/NA6
NS13 = NA4/NA5
NS14 = NA4/NA6
NS15 = NA5/NA6
NS16 = (NA1+NA2+NA3+NA4)/(NA5+NA6)
# Reflection Indices
RI1 = (ppg_window_main[ppg_maxslope] - ppg_last_min) / (ppg_window_main[ppg_window_systolic] - ppg_last_min)
RI2 = (ppg_window_main[ppg_dic] - ppg_last_min) / (ppg_window_main[ppg_window_systolic] - ppg_last_min)
RI3 = (ppg_window_main[ppg_inf] - ppg_last_min) / (ppg_window_main[ppg_window_systolic] - ppg_last_min)
RI4 = (ppg_window_main[ppg_dia] - ppg_last_min) / (ppg_window_main[ppg_window_systolic] - ppg_last_min)
# LASI
SI1 = details[i][2]/(ppg_dia-ppg_window_systolic)
SI2 = details[i][2]/(ppg_dic-ppg_window_systolic)
SI3 = details[i][2]/(ppg_inf-ppg_window_systolic)
SI4 = details[i][2]/(ppg_dia-ppg_maxslope)
SI5 = details[i][2]/(ppg_window_systolic-ppg_maxslope)
SI6 = details[i][2]/(ppg_dia-ppg_inf)
# Non-linear features
ln_mNPV1 = np.log(mNPV1)
ln_mNPV2 = np.log(mNPV2)
exp_mNPV1 = np.exp(mNPV1)
exp_mNPV2 = np.exp(mNPV2)
ln_HR = np.log(HR)
exp_HR = np.exp(HR)
ln_HRmNPV1 = np.log(mNPV1 * HR)
ln_HRmNPV2 = np.log(mNPV2 * HR)
ln_RI1 = np.log(RI1)
ln_RI2 = np.log(RI2)
ln_RI3 = np.log(RI3)
ln_RI4 = np.log(RI4)
window_features.extend([ppg_dc,ppg_ac,mNPV1,mNPV2,A1,A2,A3,A4,A5,A6,S1,S2,S3,S4,S5,S6,S7,S8,S9,S10,S11,S12,S13,S14,S15,S16,NA1,NA2,NA3,NA4,NA5,NA6,NS1,NS2,NS3,NS4,NS5,NS6,NS7,NS8,NS9,NS10,NS11,NS12,NS13,NS14,NS15,NS16,RI1
,RI2,RI3,RI4,SI1,SI2,SI3,SI4,SI5,SI6,ln_mNPV1,ln_mNPV2,exp_mNPV1,exp_mNPV2,ln_HRmNPV1,ln_HRmNPV2,ln_RI1,ln_RI2,ln_RI3,ln_RI4,HR,ln_HR,exp_HR])
window_features.extend([ppg_maxslope,ppg_window_systolic,ppg_dic,ppg_dia,ppg_inf])
window_features.extend(time_domain_features)
window_features.extend(freq_domain_features)
segment_features.append(window_features)
subject_features.append(np.mean(segment_features,0))
except:
print('^', end='')
pass
if len(np.median(subject_features,0))>0:
all_features.append(np.median(subject_features,0))
print("")
print("++")
except:
print("**")
pass
all_features = np.array(all_features)
np.savetxt(save_name, all_features, delimiter=",")